I don't think there is so much of difference between labeling and understanding things. I mean, we humans don't even understand ourselves and yet we tend to use

I don't think there is so much of difference between labeling and understanding things. I mean, we humans don't even understand ourselves and yet we tend to use our intelligence as the general measure of intelligence. But if labeling is enough, what is missing? The current approaches are mostly based on single dimensions, like pure text, images, sound etc. but rarely a combination of them. So when a human, who lives in a world where images, sounds, physical experiences, emotions and abstract information have to be combined meets an AI that lives in a single dimension it is easy to see the blind spots. So maybe just giving machines access to all the dimensions we humans perceive might help to let them act more like we do.